print('\n')
# prepare sink
gridstr = '' if grid is 'NARR' else '_'+grid
filename = avgfile.format(gridstr,'_'+periodstr)
if os.path.exists(avgfolder+filename): os.remove(avgfolder+filename)
sink = DatasetNetCDF(name='NARR Climatology', folder=avgfolder, filelist=[filename], atts=source.atts, mode='w')
sink.atts.period = periodstr
# determine averaging interval
offset = source.time.getIndex(period[0]-1979)/12 # origin of monthly time-series is at January 1979
# initialize processing
# CPU = CentralProcessingUnit(source, sink, varlist=['precip', 'T2'], tmp=True) # no need for lat/lon
CPU = CentralProcessingUnit(source, sink, varlist=None, tmp=True) # no need for lat/lon
# start processing climatology
CPU.Climatology(period=period[1]-period[0], offset=offset, flush=False)
# sync temporary storage with output
CPU.sync(flush=True)
# # make new masks
# sink.mask(sink.landmask, maskSelf=False, varlist=['snow','snowh','zs'], invert=True, merge=False)
# add names and length of months
sink.axisAnnotation('name_of_month', name_of_month, 'time',
atts=dict(name='name_of_month', units='', long_name='Name of the Month'))
#print ' === month === '
# sink += VarNC(sink.dataset, name='length_of_month', units='days', axes=(sink.time,), data=days_per_month,
# atts=dict(name='length_of_month',units='days',long_name='Length of Month'))
# close...
def computeClimatology(experiment,
filetype,
domain,
periods=None,
offset=0,
griddef=None,
varlist=None,
ldebug=False,
loverwrite=False,
lparallel=False,
pidstr='',
logger=None):
''' worker function to compute climatologies for given file parameters. '''
# input type checks
if not isinstance(experiment, Exp): raise TypeError
if not isinstance(filetype, basestring): raise TypeError
if not isinstance(domain, (np.integer, int)): raise TypeError
if periods is not None and not (isinstance(periods, (tuple, list))
and isInt(periods)):
raise TypeError
if not isinstance(offset, (np.integer, int)): raise TypeError
if not isinstance(loverwrite, (bool, np.bool)): raise TypeError
if griddef is not None and not isinstance(griddef, GridDefinition):
raise TypeError
#if pidstr == '[proc01]': raise TypeError # to test error handling
# load source
dataset_name = experiment.name
fileclass = fileclasses[filetype] # used for target file name
tsfile = fileclass.tsfile.format(domain, '')
expfolder = experiment.avgfolder
filepath = '{:s}/{:s}'.format(expfolder, tsfile)
logger.info('\n\n{0:s} *** Processing Experiment {1:<15s} *** '.
format(pidstr, "'{:s}'".format(dataset_name)) +
'\n{0:s} *** {1:^37s} *** \n'.format(
pidstr, "'{:s}'".format(tsfile)))
# check file and read begin/enddates
if not os.path.exists(filepath):
#raise IOError, "Source file '{:s}' does not exist!".format(filepath)
# print message and skip
skipmsg = "\n{:s} >>> File '{:s}' in dataset '{:s}' is missing --- skipping!".format(
pidstr, tsfile, dataset_name)
skipmsg += "\n{:s} >>> ('{:s}')\n".format(pidstr, filepath)
logger.warning(skipmsg)
# N.B.: this can cause a lot of error messages, when not all files are present
else: # if monthly source file exists
import netCDF4 as nc
ncfile = nc.Dataset(filepath, mode='r')
begintuple = ncfile.begin_date.split('-')
endtuple = ncfile.end_date.split('-')
ncfile.close()
# N.B.: at this point we don't want to initialize a full GDAL-enabled dataset, since we don't even
# know if we need it, and it creates a lot of overhead
# determine age of source file
if not loverwrite:
sourceage = datetime.fromtimestamp(os.path.getmtime(filepath))
# figure out start date
filebegin = int(begintuple[0]) # first element is the year
fileend = int(endtuple[0]) # first element is the year
begindate = offset + filebegin
if not (filebegin <= begindate <= fileend): raise DateError
# handle cases where the first month in the record is not January
firstmonth = int(begintuple[1]) # second element is the month
shift = firstmonth - 1 # will be zero for January (01)
## loop over periods
if periods is None: periods = [begindate - fileend]
# periods.sort(reverse=True) # reverse, so that largest chunk is done first
source = None # will later be assigned to the source dataset
for period in periods:
# figure out period
enddate = begindate + period
if filebegin > enddate:
raise DateError, 'End date earlier than begin date.'
if enddate - 1 > fileend: # if filebegin is 1979 and the simulation is 10 years, fileend will be 1988, not 1989!
# if end date is not available, skip period
endmsg = "\n{:s} --- Invalid Period for '{:s}': End Date {:4d} not in File! --- \n".format(
pidstr, dataset_name, enddate)
endmsg += "{:s} --- ('{:s}')\n".format(pidstr, filepath)
logger.info(endmsg)
else: ## perform averaging for selected period
# determine if sink file already exists, and what to do about it
periodstr = '{0:4d}-{1:4d}'.format(begindate, enddate)
gridstr = '' if griddef is None or griddef.name is 'WRF' else '_' + griddef.name
filename = fileclass.climfile.format(domain, gridstr,
'_' + periodstr)
if ldebug: filename = 'test_' + filename
if lparallel: tmppfx = 'tmp_wrfavg_{:s}_'.format(pidstr[1:-1])
else: tmppfx = 'tmp_wrfavg_'.format(pidstr[1:-1])
tmpfilename = tmppfx + filename
assert os.path.exists(expfolder)
filepath = expfolder + filename
tmpfilepath = expfolder + tmpfilename
lskip = False # else just go ahead
if os.path.exists(filepath):
if not loverwrite:
age = datetime.fromtimestamp(
os.path.getmtime(filepath))
# if sink file is newer than source file, skip (do not recompute)
if age > sourceage and os.path.getsize(filepath) > 1e6:
lskip = True
# N.B.: NetCDF files smaller than 1MB are usually incomplete header fragments from a previous crash
#print sourceage, age
if not lskip: os.remove(filepath)
# depending on last modification time of file or overwrite setting, start computation, or skip
if lskip:
# print message
skipmsg = "\n{:s} >>> Skipping: file '{:s}' in dataset '{:s}' already exists and is newer than source file.".format(
pidstr, filename, dataset_name)
skipmsg += "\n{:s} >>> ('{:s}')\n".format(
pidstr, filepath)
logger.info(skipmsg)
else:
if griddef is None: lregrid = False
else: lregrid = True
## begin actual computation
beginmsg = "\n{:s} <<< Computing '{:s}' (d{:02d}) Climatology from {:s}".format(
pidstr, dataset_name, domain, periodstr)
if not lregrid: beginmsg += " >>> \n"
else:
beginmsg += " ('{:s}' grid) >>> \n".format(
griddef.name)
logger.info(beginmsg)
## actually load datasets
if source is None:
source = loadWRF_TS(
experiment=experiment,
filetypes=[filetype],
domains=domain) # comes out as a tuple...
if not lparallel and ldebug:
logger.info('\n' + str(source) + '\n')
# prepare sink
if os.path.exists(tmpfilepath):
os.remove(tmpfilepath) # remove old temp files
sink = DatasetNetCDF(name='WRF Climatology',
folder=expfolder,
filelist=[tmpfilename],
atts=source.atts.copy(),
mode='w')
sink.atts.period = periodstr
# if lregrid: addGDALtoDataset(sink, griddef=griddef)
# initialize processing
CPU = CentralProcessingUnit(
source,
sink,
varlist=varlist,
tmp=lregrid,
feedback=ldebug) # no need for lat/lon
# start processing climatology
if shift != 0:
logger.info(
'{0:s} (shifting climatology by {1:d} month, to start with January) \n'
.format(pidstr, shift))
CPU.Climatology(period=period,
offset=offset,
shift=shift,
flush=False)
# N.B.: immediate flushing should not be necessary for climatologies, since they are much smaller!
# reproject and resample (regrid) dataset
if lregrid:
CPU.Regrid(griddef=griddef, flush=True)
logger.info('{:s} --- {:s} --- \n'.format(
pidstr, griddef.name))
logger.debug('{:s} --- {:s} --- \n'.format(
pidstr, str(griddef)))
# sync temporary storage with output dataset (sink)
CPU.sync(flush=True)
# add Geopotential Height Variance
if 'GHT_Var' in sink and 'Z_var' not in sink:
data_array = (sink['GHT_Var'].data_array -
sink['Z'].data_array**2)**0.5
atts = dict(
name='Z_var',
units='m',
long_name=
'Square Root of Geopotential Height Variance')
sink += Variable(axes=sink['Z'].axes,
data=data_array,
atts=atts)
# add (relative) Vorticity Variance
if 'Vorticity_Var' in sink and 'zeta_var' not in sink:
data_array = (sink['Vorticity_Var'].data_array -
sink['zeta'].data_array**2)**0.5
atts = dict(
name='zeta_var',
units='1/s',
long_name=
'Square Root of Relative Vorticity Variance')
sink += Variable(axes=sink['zeta'].axes,
data=data_array,
atts=atts)
# add names and length of months
sink.axisAnnotation('name_of_month',
name_of_month,
'time',
atts=dict(
name='name_of_month',
units='',
long_name='Name of the Month'))
if not sink.hasVariable('length_of_month'):
sink += Variable(name='length_of_month',
units='days',
axes=(sink.time, ),
data=days_per_month,
atts=dict(
name='length_of_month',
units='days',
long_name='Length of Month'))
# close... and write results to file
sink.sync()
sink.close()
writemsg = "\n{:s} >>> Writing to file '{:s}' in dataset {:s}".format(
pidstr, filename, dataset_name)
writemsg += "\n{:s} >>> ('{:s}')\n".format(
pidstr, filepath)
logger.info(writemsg)
# rename file to proper name
if os.path.exists(filepath):
os.remove(filepath) # remove old file
os.rename(tmpfilepath,
filepath) # this will overwrite the old file
# print dataset
if not lparallel and ldebug:
logger.info('\n' + str(sink) + '\n')
# clean up (not sure if this is necessary, but there seems to be a memory leak...
del sink, CPU
gc.collect() # get rid of these guys immediately
# clean up and return
if source is not None:
source.unload()
del source
# N.B.: source is only loaded once for all periods
# N.B.: garbage is collected in multi-processing wrapper as well
# return
return 0 # so far, there is no measure of success, hence, if there is no crash...